[cod] Linux beta server problems

Mark J. DeFilippis defilm at acm.org
Sat Oct 29 19:41:54 EDT 2005


Yes, Sorry. I was rushing for dinner, and thanks for making the corrections.

Indeed, the Tyans are of course are a HyperTrasport MOBO. The CPU's are 
Dual Core Opterons,
hence 8 CPU's, and do not support HyperThreading as you correctly stated.
(Although this would make the Opteron a bit more practical for the client 
side if it did, as dual core
don't do much for the client side as you know. This shows up as 8 CPU's. 
(Not to mention that the
client side application would have to support the dual core processor and I 
don't see that happening
any time soon, which is not to say it won't.

The DUal Intel P4 is HypterThreading, and the Intel based MOBO's do not 
support HyperTransport
as you correctly point out. This of course, as you pointed out shows up as 
4 CPU's, 2 which are of
course real, the other two are virtual. :)

However, I will say this much for the 3.2 P4 Xeon HyperThread procs, using 
2.4 with 2.6SMP yields
much better results in processor thread scheduling than it did on 2.3 with 
the SMP kernel.
So nice gain on the same hardware by just upgrading from 2.3 to 2.4 with 
2.6SMP... If anyone
is running this out there and 2.4 with 2.6SMP is an option, go for it. You 
WILL notice.

As much as I admit the Tyan HyperTransport motherboard with Dual Opteron's is a
powerhouse, it is not my private gaming machine for me and my boys ;-)  It is a
work machine, (but I have fun stuff running on it, and I won't give it back 
ever) :)

I widely use it to do Queuing, modeling, and modeling of stochastic processes I
am a high speed transport network architect by trade.  This chews up tons 
of CPU.
Some of the models with data collected would run several days if it were 
not for this puppy.
But when it is not doing work, it is all mine. :)

(As a side note, honestly the logical CPU of the HyperTransport board based 
Opteron does
not yield the same performance as the primary processor. So although real
and logical, it at times behaves virtual.  Where if I had 8 real processors 
in a
grid, I could truly schedule processes in multiple ways to the processors. 
Don't
get the ability with the Dual core Opterons.

Thanks for the corrections.

Mark


At 05:59 PM 10/29/2005, you wrote:

>I think you're mistaken, as this is a HyperTransport board, not 
>Hyperthreading (hyperthreading is an Intel technology, HyperTransport is 
>an AMD technology.. and they are very different techs at that). They 
>aren't virtual cpus, but logical cpus.
>
>Intel P4/Xeons use Hyperthreading (which show up as logical cpus as well).
>Opteron based boards are HyperTransport, and dont support Hyperthreading.
>
>It sounds like you have 4 dual core processors (so you have a total of 8 
>Logical, but very real processors). Very nice rig.. Wish it were mine :)
>
>A Quad Dual Core Xeon would show up w/ 16 logical processors (8 very real 
>processors and 8 very virtual hyperthreading units).
>
>Just don't want anyone getting confused!

S1,-------------------------------------------------------------------------------
Mark J. DeFilippis, Ph. D EE          defilm at acm.org
                                       defilm at ieee.org


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://icculus.org/pipermail/cod/attachments/20051029/98337b1b/attachment.htm>


More information about the Cod mailing list